A high performance hybrid LSTM CNN secure architecture for IoT environments using deep learning

Abstract
The growing use of IoT has brought enormous safety issues that constantly demand stronger hide from increasing risks of intrusions. This paper proposes an Advanced LSTM-CNN Secure Framework to optimize real-time intrusion detection in the IoT context. It adds LSTM layers, which allow for temporal dependencies to be learned, and CNN layers to decompose spatial features which makes this model efficient in identifying threats. It is important to note that the used BoT-IoT dataset involves various cyber attack typologies like DDoS, botnet, reconnaissance, and data exfiltration. These outcomes present that the proposed LSTM-CNN model has 99.87% accuracy, 99.89% precision, and 99.85% recall with a low false positive rate of 0.13% and exceeds CNN, RNN, Standard LSTM, BiLSTM, GRU deep learning models. In addition, the model has 90.2% accuracy in conditions of adversarial attack proving that the model is robust and can be used for practical purposes. Based on feature importance analysis using SHAP, the work finds that packet size, connection duration, and protocol type should be the possible indicators for threat detection. These outcomes suggest that the Hybrid LSTM-CNN model could be useful in improving the security of IoT devices to provide increased reliability with low false alarm rates.